- 09 Jun, 2023 40 commits
-
-
Zhangjin Wu authored
These 2 test cases are added to cover the normal using scenes of gettimeofday(). They have been used to trigger and fix up such issue with nolibc: nolibc-test.c:(.text.gettimeofday+0x54): undefined reference to `__aeabi_ldivmod' This issue happens while there is no "unsigned int" conversion in the coming new clock_gettime / clock_gettime64 syscall path of gettimeofday(): tv->tv_usec = ts.tv_nsec / 1000; Suggested-by:Thomas Weißschuh <linux@weissschuh.net> Link: https://lore.kernel.org/linux-riscv/280867a8-7601-4a96-9b85-87668e1f1282@t-8ch.de/Signed-off-by:
Zhangjin Wu <falcon@tinylab.org> Reviewed-by:
Willy Tarreau <w@1wt.eu> Signed-off-by:
Paul E. McKenney <paulmck@kernel.org>
-
Zhangjin Wu authored
In the clock_gettime / clock_gettime64 syscalls based gettimeofday(), there is no way to let kernel space 'fixup' the invalid data pointer of 'struct timeval' and 'struct timezone' for us for we need to read timespec from kernel space and then convert to timeval in user-space ourselves and also we need to simply ignore and reset timezone in user-space. Without this removal, the invalid (void *)1 address will trigger a sigsegv (signum = 11) signal and stop the whole test. Suggested-by:
-
Zhangjin Wu authored
Some functions may be implemented with different syscalls in different platforms, these syscalls may set different errnos for the same arguments, let's support such cases. Suggested-by:
-
Zhangjin Wu authored
In function ‘open’: nolibc/sysroot/arm/include/sys.h:919:23: warning: ‘mode_t’ {aka ‘short unsigned int’} is promoted to ‘int’ when passed through ‘...’ 919 | mode = va_arg(args, mode_t); | ^ nolibc/sysroot/arm/include/sys.h:919:23: note: (so you should pass ‘int’ not ‘mode_t’ {aka ‘short unsigned int’} to ‘va_arg’) nolibc/sysroot/arm/include/sys.h:919:23: note: if this code is reached, the program will abort Signed-off-by: -
Zhangjin Wu authored
This is required by the coming removal of the oldselect and newselect support. pselect6/pselect6_time64 will be used unconditionally, they have 6 arguments. Suggested-by:
Arnd Bergmann <arnd@arndb.de> Link: https://lore.kernel.org/linux-riscv/bf3e07c1-75f5-425b-9124-f3f2b230e63a@app.fastmail.com/Signed-off-by:
-
Zhangjin Wu authored
nolibc now has INT_MAX in stdint.h, so, don't mix INT_MAX and __INT_MAX__, unify them to INT_MAX. Signed-off-by:
-
Zhangjin Wu authored
When compile nolibc-test.c with 2.31 glibc, we got such error: In file included from /usr/riscv64-linux-gnu/include/sys/cdefs.h:452, from /usr/riscv64-linux-gnu/include/features.h:461, from /usr/riscv64-linux-gnu/include/bits/libc-header-start.h:33, from /usr/riscv64-linux-gnu/include/limits.h:26, from /usr/lib/gcc-cross/riscv64-linux-gnu/9/include/limits.h:194, from /usr/lib/gcc-cross/riscv64-linux-gnu/9/include/syslimits.h:7, from /usr/lib/gcc-cross/riscv64-linux-gnu/9/include/limits.h:34, from /labs/linux-lab/src/linux-stable/tools/testing/selftests/nolibc/nolibc-test.c:6: /usr/riscv64-linux-gnu/include/bits/wordsize.h:28:3: error: #error "rv32i-based targets are not supported" 28 | # error "rv32i-based targets are not supported" Glibc (>= 2.33) commit 5b6113d62efa ("RISC-V: Support the 32-bit ABI implementation") fixed up above error. As suggested by Thomas, defining INT_MIN/INT_MAX for nolibc can remove the including of limits.h, and therefore no above error. of course, the other libcs still require limits.h, move it to the right place. The LONG_MIN/LONG_MAX are also defined too. Suggested-by: -
Zhangjin Wu authored
Compiling nolibc-test.c with gcc on x86_64 got such warning: tools/testing/selftests/nolibc/nolibc-test.c: In function ‘expect_eq’: tools/testing/selftests/nolibc/nolibc-test.c:177:24: warning: format ‘%lld’ expects argument of type ‘long long int’, but argument 2 has type ‘uint64_t’ {aka ‘long unsigned int’} [-Wformat=] 177 | llen += printf(" = %lld ", expr); | ~~~^ ~~~~ | | | | | uint64_t {aka long unsigned int} | long long int | %ld It because that glibc defines uint64_t as "unsigned long int" when word size (means sizeof(long)) is 64bit (see include/bits/types.h), but nolibc directly use the 64bit "unsigned long long" (see tools/include/nolibc/stdint.h), which is simpler, seems kernel uses it too (include/uapi/asm-generic/int-ll64.h). use a simple conversion to solve it. Suggested-by: -
Zhangjin Wu authored
The opensbi package from Ubuntu 20.04 only provides rv64 firmwares: $ dpkg -S opensbi | grep -E "fw_.*bin|fw_.*elf" | uniq opensbi: /usr/lib/riscv64-linux-gnu/opensbi/generic/fw_dynamic.bin opensbi: /usr/lib/riscv64-linux-gnu/opensbi/generic/fw_jump.bin opensbi: /usr/lib/riscv64-linux-gnu/opensbi/generic/fw_dynamic.elf opensbi: /usr/lib/riscv64-linux-gnu/opensbi/generic/fw_jump.elf To run this nolibc test for rv32, users must build opensbi or download a prebuilt one from qemu repository: https://gitlab.com/qemu-project/qemu/-/blob/master/pc-bios/opensbi-riscv32-generic-fw_dynamic.bin And then use -bios to tell qemu use it to avoid such failure: $ qemu-system-riscv32 -display none -no-reboot -kernel /path/to/arch/riscv/boot/Image -serial stdio -M virt -append "console=ttyS0 panic=-1" qemu-system-riscv32: Unable to load the RISC-V firmware "opensbi-riscv32-generic-fw_dynamic.bin" To run from makefile, QEMU_ARGS_EXTRA is added to allow pass extra arguments like -bios: $ make run QEMU_ARGS_EXTRA="-bios /path/to/opensbi-riscv32-generic-fw_dynamic.bin" ... Suggested-by: -
Thomas Weißschuh authored
gettimeofday() is not guaranteed by posix to handle a NULL value as first argument gracefully. On glibc for example it crashes. (When not going through the vdso) Link: https://lore.kernel.org/lkml/96f1134d-ce6e-4d82-ae00-1cd4038809c4@t-8ch.de/Signed-off-by:
-
Thomas Weißschuh authored
On 32bit platforms size_t is not enough to represent [u]int_fast64_t. Fixes: 3e9fd4e9 ("tools/nolibc: add integer types and integer limit macros") Signed-off-by:
-
Zhangjin Wu authored
running nolibc-test with glibc on x86_64 got such print issue: 29 execve_root = -1 EACCES [OK] 30 fork30 fork = 0 [OK] 31 getdents64_root = 712 [OK] The fork test case has three printf calls: (1) llen += printf("%d %s", test, #name); (2) llen += printf(" = %d %s ", expr, errorname(errno)); (3) llen += pad_spc(llen, 64, "[FAIL]\n"); --> vfprintf() In the following scene, the above issue happens: (a) The parent calls (1) (b) The parent calls fork() (c) The child runs and shares the print buffer of (1) (d) The child exits, flushs the print buffer and closes its own stdout/stderr * "30 fork" is printed at the first time. (e) The parent calls (2) and (3), with "\n" in (3), it flushs the whole buffer * "30 fork = 0 ..." is printed Therefore, there are two "30 fork" in the stdout. Between (a) and (b), if flush the stdout (and the sterr), the child in stage (c) will not be able to 'see' the print buffer. Signed-off-by: -
Zhangjin Wu authored
The ppoll and ppoll_time64 syscalls have 5 arguments, but we only provide 4, align with kernel and add the missing sigsetsize argument. Because the sigmask is NULL, the last sigsetsize argument is ignored, keep it as 0 here is safe enough. Signed-off-by:
-
Zhangjin Wu authored
There were two exactly similar occurrences of this test. Signed-off-by:
-
Zhangjin Wu authored
EOVERFLOW will be used in the coming time64 syscalls support. Signed-off-by:
-
Thomas Weißschuh authored
Keep backwards compatibility through unions. The compatibility macros like #define st_atime st_atim.tv_sec as documented in stat(3type) don't work for nolibc because it would break with other stat-like structures that contain the field st_atime. The stx_atime, stx_mtime, stx_ctime are in type of 'struct statx_timestamp', which is incompatible with 'struct timespec', should be converted explicitly. /* include/uapi/linux/stat.h */ struct statx_timestamp { __s64 tv_sec; __u32 tv_nsec; __s32 __reserved; }; /* include/uapi/linux/time.h */ struct timespec { __kernel_old_time_t tv_sec; /* seconds */ long tv_nsec; /* nanoseconds */ }; Signed-off-by: -
Thomas Weißschuh authored
The child process forked during stackprotector tests intentionally gets killed with SIGABRT. By default this will trigger writing a coredump. The writing of the coredump can spam the systems coredump machinery and take some time. Timings for the full run of nolibc-test: Before: 200ms After: 20ms This is on a desktop x86 system with systemd-coredumpd enabled. Signed-off-by:
-
Thomas Weißschuh authored
It will be used to disable core dumps from the child spawned to validate the stack protector functionality. Signed-off-by:
-
Thomas Weißschuh authored
s390 does not support the "global" stack protector mode that is implemented in nolibc. Now that nolibc detects if stack protectors are enabled at runtime it could happen that a future compiler does indeed use global mode on and nolibc would compile but segfault at runtime. To avoid this hypothetic case and to align s390 with the other architectures disable stack protectors when compiling _start(). Signed-off-by:
-
Thomas Weißschuh authored
Not all compilers, notably GCC < 10, have support for __attribute__((no_stack_protector)). Fall back to a mechanism that also works there. Tested with GCC 9.5.0 from kernel.org crosstools. Signed-off-by:
-
Thomas Weißschuh authored
Now that nolibc enable stackprotector support automatically when the compiler enables it we only have to get the -fstack-protector flags correct. The cc-options are structured so that -fstack-protector-all is only enabled if -mstack-protector=guard works, as that is the only mode supported by nolibc. Signed-off-by:
-
Thomas Weißschuh authored
The stackprotector support in nolibc should be enabled iff it is also enabled in the compiler. Use the preprocessor defines added by gcc and clang if stackprotector support is enable to automatically do so in nolibc. This completely removes the need for any user-visible API. To avoid inlining the lengthy preprocessor check into every user introduce a new header compiler.h that abstracts the logic away. As the define NOLIBC_STACKPROTECTOR is now not user-relevant anymore prefix it with an underscore. Suggested-by:
-
Thomas Weißschuh authored
This makes it easier to add and remove more entries in the future without creating spurious diff hunks. Signed-off-by:
-
Thomas Weißschuh authored
Signed-off-by:
-
Thomas Weißschuh authored
The all-zero pattern is one of the more probable out-of-bound writes so add a special case to not accidentally accept it. Also it enables the reliable detection of stack protector initialization during testing. Signed-off-by:
-
Thomas Weißschuh authored
This was forgotten in the original submission. It is unknown why it worked for x86_64 on some compiler without this attribute. Reported-by:
-
Thomas Weißschuh authored
Signed-off-by:
-
Thomas Weißschuh authored
Signed-off-by:
Palmer Dabbelt <palmer@rivosinc.com> Signed-off-by:
-
Thomas Weißschuh authored
Signed-off-by:
-
Thomas Weißschuh authored
Signed-off-by:
-
Thomas Weißschuh authored
Signed-off-by:
-
Thomas Weißschuh authored
Signed-off-by:
-
Thomas Weißschuh authored
Previously each space character used for alignment during test execution was written in a single write() call. This would make the output from strace fairly unreadable. Coalesce all spaces into a single call to write(). Signed-off-by:
-
Zhangjin Wu authored
Compiling nolibc-test.c for rv32 got such error: tools/testing/selftests/nolibc/nolibc-test.c:599:57: error: ‘__NR_fstat’ undeclared (first use in this function) 599 | CASE_TEST(syscall_args); EXPECT_SYSER(1, syscall(__NR_fstat, 0, NULL), -1, EFAULT); break; The generic include/uapi/asm-generic/unistd.h used by rv32 doesn't support __NR_fstat, use the more generic __NR_statx instead: Running test 'syscall' 69 syscall_noargs = 1 [OK] 70 syscall_args = -1 EFAULT [OK] __NR_statx has been added from v4.10: commit a528d35e ("statx: Add a system call to make enhanced file info available") It has been supported by all of the platforms since at least from v4.20. Suggested-by: -
Thomas Weißschuh authored
syscall() is used by "normal" libcs to allow users to directly call syscalls. By having the same syntax inside nolibc users can more easily write code that works with different libcs. The macro logic is adapted from systemtaps STAP_PROBEV() macro that is released in the public domain / CC0. Signed-off-by:
-
Zhangjin Wu authored
When compile nolibc application for rv32, we got such errors: nolibc/sysroot/riscv/include/arch.h:190: Error: unrecognized opcode `ld a4,0(a3)' nolibc/sysroot/riscv/include/arch.h:194: Error: unrecognized opcode `sd a3,%lo(_auxv)(a4)' nolibc/sysroot/riscv/include/arch.h:196: Error: unrecognized opcode `sd a2,%lo(environ)(a3)' Refer to arch/riscv/include/asm/asm.h and add REG_L/REG_S macros here to let rv32 uses its own lw/sw instructions. Signed-off-by:
-
Thomas Weißschuh authored
The same constants and some more have been exposed to userspace via linux/reboot.h for a long time. To avoid conflicts and trim down nolibc a bit drop the custom definitions. Signed-off-by:
-
Thomas Weißschuh authored
On s390 the arguments to clone() which is used by fork() are different than other archs. Make sure everything works correctly. Signed-off-by:
-
Thomas Weißschuh authored
On s390 the first two arguments to the clone() syscall are swapped, as documented in clone(2). Signed-off-by:
-
Thomas Weißschuh authored
To make sure no non-compatible changes are introduced accidentally validate the language standard when building the tests. Signed-off-by:
-
